Date post: | 29-Nov-2014 |
Category: |
Business |
Upload: | minethatdata |
View: | 2,227 times |
Download: | 0 times |
1Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
A Typical Multichannel Forensics ProjectA CEO’s Guide To Reducing Expenses
And Increasing Profit
Kevin HillstromPresident, MineThatData
[email protected]://minethatdata.blogspot.com
2Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
A Project … Requested By A CEO
Multichannel Forensics projects have evolved since analyzing internet customer behavior at Eddie Bauer in the late 1990s.
Today, a typical Multichannel Forensics project focuses on a series of objectives designed to identify customers who will continue to generate sales if advertising is reduced.
We begin with at least five years of purchase history, and an enthusiastic desire to learn about customer behavior!
3Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
The Data
We create a file that has one row for every item a customer has ever purchased.
We want key attributes like household_id, order_date, price, quantity, fulfillment information, purchase_channel, zip_code, the advertising channel that drove the purchase (search, e-mail, catalogs), and the advertising channel that influenced the purchase (i.e. matchback information), sku, merchandise category/department/division.
With this information, we can run our analysis!
4Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
Micro-Channels
A typical project focuses on “micro-channels”, combinations of advertising and physical channels that customers use to purchase merchandise.
For instance, a customer who orders online after clicking through an e-mail campaign that was matched-back to a catalog mailing is a customer who utilizes a “micro-channel”.
Micro-Channel = E-Mail / Catalog.
5Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
Micro-Channels
Most business-to-consumer and business-to-business brands have between ten and twenty popular micro-channels that customers use.
Our goal is to understand how customers repurchase after buying from a specific micro-channel. What does the customer do next?
Let’s look at an example: 2008 repurchase activity for customers who in 2007 shopped via e-mail without a catalog matched-back to the order.
6Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
E-Mail Alone
Our first job is to look at the repurchase rate for this segment … in this case, 29.8%.
Repurchase Rates:• 60%+ = Retention Mode.• 40% - 59% = Hybrid Mode.• 0% to 39% = Acquisition Mode.
The mode dictates the marketing strategy.
7Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
E-Mail Alone
Next, we look at the “rebuy index”.• 0% to 19% = Isolation Mode.• 20% to 49% = Equilibrium Mode.• 50%+ = Transfer Mode.
Equilibrium Mode means that customers are willing to try a different micro-channel. Transfer Mode means that customers are leaving for another micro-channel.
8Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
E-Mail Alone
Customers who shopped via e-mail (without catalog assistance) are likely to shop online via a catalog, through e-mail via a catalog, and to shop on the internet without the aid of advertising.
These are the micro-channels that a customer is likely to migrate to after purchasing via e-mail alone. We probably cannot force the customer to shop other micro-channels.
9Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
The Micro-Channel Ecosystem
Most business-to-consumer
10Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
The Micro-Channel Ecosystem
Notice that this brand acquires customers in traditional ways (often via catalog marketing).
Over time, the customer evolves, migrating to e-mail marketing.
Other customers come in via online marketing channels.
Eventually, many customers buy online, and do not need advertising to place an order. This is a good thing!!!!
11Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
The Micro-Channel Ecosystem
Most micro-channel ecosystems illustrate a path that customers take as they move from being advertising-dependent to generating organic demand.
Many customers prefer catalog advertising.
Many customers prefer e-mail marketing, or paid search.
Some customers will shop regardless whether they are marketed to, or not.
12Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
Putting The Information To Use
In most of my projects, I have to identify the customers who are most likely to require advertising in the future, and identify customers who are less likely to require advertising in the future.
Three distinct models are built to score every single customer in your database.
Here are the models!
13Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
Response Model
I use Logistic Regression to predict the likelihood of a customer purchasing again during the next twelve months.
Traditional variables (recency, frequency, monetary value), geographic variables (zip code forensics), micro-channel metrics, and merchandise preferences are entered into the equation.
Each customer is given a predicted (29.4%) likelihood of buying again in the next twelve months.
14Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
Spending Model
I use Ordinary Least Squares Regression to predict how much a customer will spend if the customer purchases again in the next twelve months.
Traditional variables (recency, frequency, monetary value), geographic variables (zip code forensics), micro-channel metrics, and merchandise preferences are entered into the equation.
Each customer is assigned a spending prediction ($150) for the next twelve months, if the customer does purchase.
15Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
Organic Model
Variants of regression (probit, hyperbolic tangent transformation) are used to predict the percentage of demand that will be generated without the aid of advertising.
We really focus on micro-channels at this stage of modeling.
Each customer receives a prediction (20%) of demand that will be “organic”, not caused by advertising, in the next twelve months.
16Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
Advertising Value
Our predictions are multiplied together, to determine the demand that will be driven by advertising:
• Response Prediction = 29.4%.• Spending Prediction = $150.00.• Organic Prediction = 20.0%.• Prediction = 0.294 * 150.00 * (1 – 0.20) = $8.82.
This customer will spend $8.82 in the next twelve months if the customer is advertised to. Now let’s review how we will use this value to determine a contact strategy.
17Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
Contact Strategy
We use a relationship that simulates diminishing returns.
Alternatively, testing data can be used to determine this relationship.
Demand By Advertising Level
Number of Catalogs
Ad
vert
isin
g D
eman
d
0 1 2 3 4 5 6 7 8 9 10 11 120123456789
101112131415
18Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
Optimal Contact Strategy
Given our relationships, we can estimate how many catalogs are “right” for any individual customer.
In this case, profit is maximized at two catalog mailings, not the six that this customer would normally receive.
19Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
The Future: A Five Year Sales Plan
20Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
The Future: A Five Year Sales Plan
The CEO will thoroughly understand the anticipated five year sales trajectory, by channel.
The CEO will thoroughly understand how customer retention, customer acquisition, and marketing spend interact to drive the business forward.
With this data, the CEO will be able to craft a strategy to grow sales and profit, or will be able to trim expense in the least damaging way.
21Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
E-Mail Marketing Strategy
Many folks are looking to develop a “targeted” e-mail marketing strategy --- one that is not terribly complicated, one that is easy to implement, one that is human driven ---not driven by automated algorithms.
So, many projects have an e-mail component --- one where each customer is assigned to a “merchandise preference”, so that the marketer can match an individual customer to a unique version of an e-mail campaign.
Merchandise purchases are very important here!
22Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
E-Mail Segmentation
I’ll assign customers into one of nine segments, for instance, based on how many merchandise divisions the customer buys from, and their merchandise preference.
23Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
E-Mail Marketing Strategy
Once we assign customers into one of nine different e-mail marketing segments, we match customers to the version of an e-mail campaign that is most similar to their interests.
A womens merchandise customer who only buys womensmerchandise should probably receive … womensmerchandise!
A mens merchandise customer who buys from many merchandise divisions could receive just about anything and be productive --- test and measure!!!
24Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
Multichannel Forensics And Profit
Most of the Multichannel Forensics projects I work on require a reduction in advertising expense coupled with demand maximization.
We do this by analyzing micro-channels, reducing catalogs to customers who generate demand organically.
We also do this by matching e-mail marketing campaigns to customers who have specific merchandise preferences.
25Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
Project Cost
Projects are based on the number of buyers who purchased from your business in the past twelve months.
• 10,000 buyers = $8,500.• 100,000 buyers = $15,300.• 1,000,000 buyers = $27,300.• 10,000,000 buyers = $49,000.
Many larger companies have the resources to do this work on their own. About three dozen brands have worked with me on Multichannel Forensics projects of this nature during the past two years.
26Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
Too Expensive? How Does Free Sound?
For some folks, a Multichannel Forensics project is too expensive.
So why not leverage a free tool? Zip Code Forensics is a free tool that allows you to target customers dedicated to e-commerce or catalog marketing.
27Kevin Hillstrom, President: MineThatData
Visit The Blog http://minethatdata.blogspot.com
Interested?
Most projects take about four weeks to complete, once data is received.
Multichannel Forensics projects have been completed for more than thirty marketers … from $10,000,000 online businesses to $50,000,000 catalogers to multi-billion dollar retail multi-channel brands. You get the benefit of this aggregated learning by working on a project.
Send me an e-mail ([email protected]) for project details, timelines, costs, and requirements.